Design and Implementation of a Multi-Agent Image Preprocessing Framework for Deep Learning

• Yago Salerno Dayub Campos UERJ
• Thiago de Medeiros Rocha UERJ
• Gilson Alexandre Ostwald Pedro da Costa UERJ
• Guilherme Lucio Abelha Mota UERJ
• Vera Maria Benjamim Werneck UERJ
• Tassio Ferenzini Martins Sirqueira UERJ

DOI: https://doi.org/10.5753/wesaac.2026.31126

Resumo

Artificial Intelligence techniques, such as neural networks and vision transformers, have become the preferable approaches in computer vision. However, training these algorithms depends on high-quality data, which makes preprocessing, image crop, resize, normalization, training and test sets split among others, a fundamental stage. Preprocessing is often time-consuming and complex task. Thus, this work aims at contributing to optimize this process. The idea is substituting the manual work by a multi-agent system capable of performing various preprocessing algorithms on image datasets. The proposed method is based on a modular and expandable architecture, in which specialized agents execute actions in an isolated and organized manner. The implementation demonstrates the feasibility of the proposed architecture, providing a modular and extensible framework for image preprocessing. The proposed solution facilitates the integration of new preprocessing techniques while preserving a low-coupling architecture. The herein presented experiments have shown that the proposed architecture provides an efficient framework for integrating the desired preprocessing functionalities, facilitating the development of neural networks training pipelines.

Referências

Agustsson, E. and Timofte, R. (2017). Ntire 2017 challenge on single image super-resolution: Dataset and study. In The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops.

DeVries, T. and Taylor, G. W. (2017). Improved regularization of convolutional neural networks with cutout. arXiv preprint arXiv:1708.04552.

Ioffe, S. and Szegedy, C. (2015). Batch normalization: Accelerating deep network training by reducing internal covariate shift. In International Conference on Machine Learning (ICML), pages 448–456. PMLR.

Krizhevsky, A., Sutskever, I., and Hinton, G. E. (2012). Imagenet classification with deep convolutional neural networks. In Advances in Neural Information Processing Systems (NeurIPS), volume 25.

Kumari, N., Sharma, P., and Kansal, I. (2023). An analytical review on image enhancement techniques. In 2023 2nd International Conference on Disruptive Technologies in Multi-disciplinary Research (DELCON), pages 1–6. IEEE.

Padgham, L. and Winikoff, M. (2004). Developing Intelligent Agent Systems: A Practical Guide. Wiley Series in Agent Technology. John Wiley & Sons, Chichester, UK.

Shorten, C. and Khoshgoftaar, T. M. (2019). A survey on image data augmentation for deep learning. Journal of Big Data, 6(1):1–48.

Soekarta, R. and Ku-Mahamud, K. R. (2025). Recent facial image preprocessing techniques: A review. Engineering Proceedings, 84(1):39.

Sun, H., Bian, H., Zeng, S., Rao, Y., Xu, X., Mei, L., and Gou, J. (2025). Datasetagent: A novel multi-agent system for auto-constructing datasets from real-world images. arXiv preprint arXiv:2507.08648.

Tachibana, Y., Obata, T., Kershaw, J., Sakaki, H., Urushihata, T., Omatsu, T., Kishimoto, R., and Higashi, T. (2019). The utility of applying various image preprocessing strategies to reduce the ambiguity in deep learning-based clinical image diagnosis. Magnetic Resonance in Medical Sciences, 19:92–98.